← Back to Home
Credit Risk

ECL Engine & Scorecard Modelling

A fully transparent Expected Credit Loss engine built for IndAS 109 and IFRS 9 compliance — covering all segments with statistically rigorous PD, LGD, and EAD models that pass regulatory and audit scrutiny.

ECL Engine Walkthrough

Modelling Capabilities

Built from first principles by statisticians who have modelled credit risk inside regulated institutions.

PD, LGD & EAD Modelling

Probability of Default, Loss Given Default, and Exposure at Default models Vasicek PD Model, Pluto Tache LDP adjustments — calibrated to Indian portfolio data.

Vintage Analysis & Migration Matrices

Cohort-based vintage analysis and transition matrix construction across loan tenors and borrower segments, providing the statistical foundation for stage allocation and provision estimation.

Forward-looking Macro Adjustments

Point-in-time adjustments overlaying RBI macroeconomic scenarios onto through-the-cycle estimates, ensuring provisioning reflects current economic conditions and regulatory expectations.

Full Segment Coverage

Purpose-built models for Retail, SME, Corporate, and Microfinance portfolios — each with segment-specific behavioural assumptions, collateral treatment, and regulatory classification rules.

Statistical Scorecard Modelling

Origination and behavioural scorecards built using rigorous variable selection, WoE binning, and IV analysis — fully documented for model governance and validation committees.

Audit-ready Transparency

Every calculation step is fully traceable. Model documentation, assumption registers, and back-testing reports are generated automatically — designed to withstand Big 4 and RBI scrutiny.

"The transparency and statistical rigor of the Sheep.AI ECL engine passed our Big 4 audit without a single qualification."

Chief Risk Officer — Small Finance Bank

Sourced live from RBI DBIE

Macroeconomic Inputs

Forward-looking ECL under IndAS 109 requires more than a single point-in-time adjustment — it requires a consistent, jointly-forecast macro environment. Every quarter, the engine pulls the underlying series directly from RBI's Database on Indian Economy (DBIE), so the macro overlay is always anchored to the regulator's own published data rather than a third-party proxy.

Repo Rate

RBI's policy rate, the primary transmission channel into lending rates and borrower repayment capacity.

CPI Inflation

Headline and core CPI series, driving real income erosion assumptions in retail PD models.

Real GDP Growth

Quarterly GVA and GDP growth rates, anchoring the macro cycle position for corporate and SME segments.

Index of Industrial Production

Sectoral IIP series used to capture industry-specific stress in corporate exposure portfolios.

Gross NPA Ratio (SCBs)

System-wide asset quality trends from RBI's banking sector data, used to benchmark portfolio-level PD calibration.

Unemployment Rate (PLFS)

Periodic Labour Force Survey unemployment data, a key behavioural driver for retail and microfinance default risk.

USD/INR Exchange Rate

Currency series relevant to import-linked corporate borrowers and external commercial borrowing exposure.

Bank Credit & Deposit Growth

Aggregate credit and deposit growth rates, reflecting system liquidity and credit appetite over the cycle.

Forecasting Methodology

Cointegrated VAR Forecasting, 40 Quarters Out

Macro variables don't move independently — rates, inflation and growth are bound together by long-run equilibrium relationships. Forecasting them one at a time produces paths that drift apart and break those relationships. A Cointegrated Vector Autoregression — estimated as a Vector Error Correction Model (VECM) — forecasts the entire macro system jointly, so the 10-year (40-quarter) path used for lifetime PD stays internally consistent.

01

Data Ingestion from RBI DBIE

Quarterly time series for every macro factor are pulled directly from RBI's Database on Indian Economy (DBIE), the same source RBI itself publishes for monetary and financial statistics.

02

Stationarity Testing

Each series is tested using Augmented Dickey-Fuller and KPSS tests. Indian macro series are typically I(1) — non-stationary in levels but stationary in first differences.

03

Johansen Cointegration Test

Because the variables move together over the long run (e.g. inflation, rates and growth), the Johansen procedure identifies how many cointegrating vectors — stable long-run equilibrium relationships — exist among them.

04

VECM Estimation

A Vector Error Correction Model is fit jointly across all macro variables, capturing both short-run shocks and the speed at which each variable reverts to long-run equilibrium after a deviation.

05

40-Quarter Forecast & Scenario Weighting

The fitted VECM is iterated forward 40 quarters — a full 10-year horizon — under Base, Upside and Downside paths calibrated to RBI's published macro stress scenarios, feeding directly into the lifetime PD term structure.

40 Qtrs

10-year forecast horizon, rebuilt every quarter

8 Variables

Forecast jointly, not in isolation

3 Scenarios

Base, Upside and Downside, mapped to RBI stress paths

IndAS 109 / IFRS 9

The ECL Framework, Stage by Stage

IndAS 109 (aligned with IFRS 9) replaces the incurred-loss model with an expected-loss model. Every exposure is classified into one of three stages based on how much credit risk has changed since origination, and the stage determines whether provisioning covers 12 months or the full remaining life of the loan.

Stage 112-month ECL

Performing

No significant increase in credit risk since origination. Provisioning is based on the probability of default occurring within the next 12 months.

Stage 2Lifetime ECL

Significant Increase in Credit Risk (SICR)

Triggered by DPD thresholds, internal rating downgrades, or qualitative overlays. Provisioning moves to expected losses over the full remaining life of the exposure.

Stage 3Lifetime ECL

Credit-impaired

Objective evidence of default (typically DPD > 90). Lifetime ECL is recognised and interest income is computed on the net carrying amount.

How a single ECL number is built

For every exposure, the engine projects PD, LGD and EAD across each future period, discounts the resulting expected loss back to the reporting date at the original effective interest rate (EIR), and sums the result over either 12 months (Stage 1) or the full remaining tenor (Stage 2 and 3).

The forward-looking macro paths from the cointegrated VAR feed directly into the PD term structure, and the Base, Upside and Downside scenario outcomes are combined into a single probability-weighted expected value — not just the most-likely path — as required under IndAS 109's forward-looking measurement objective.

Significant Increase in Credit Risk (SICR) — the Stage 1 to Stage 2 trigger — is assessed using a combination of days-past-due thresholds, internal or external rating migration, and qualitative overlays, each independently configurable and fully documented for validation review.

IndAS 109

Full regulatory compliance out of the box

4 Segments

Retail, SME, Corporate & Microfinance

Big 4 Validated

Audit-ready documentation and model trails

See the engine running on live data

Our live demo environment is pre-loaded with anonymised portfolio data so you can explore every calculation step in real time.